未验证 提交 83035e9d 编写于 作者: W Wu Yi 提交者: GitHub

Merge pull request #13382 from typhoonzero/fix_fluid_benchmark_resnet_lr

fix fluid_benchmark resnet lr decay
...@@ -207,7 +207,7 @@ def get_model(args, is_train, main_prog, startup_prog): ...@@ -207,7 +207,7 @@ def get_model(args, is_train, main_prog, startup_prog):
total_images = 1281167 / trainer_count total_images = 1281167 / trainer_count
step = int(total_images / args.batch_size + 1) step = int(total_images / (args.batch_size * args.gpus) + 1)
epochs = [30, 60, 90] epochs = [30, 60, 90]
bd = [step * e for e in epochs] bd = [step * e for e in epochs]
base_lr = args.learning_rate base_lr = args.learning_rate
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册